Production control methods and systems for aloe vera beverages

Abstract

This application relates to the field of food production control technology, and discloses a production control method and system for aloe vera beverages. The method includes: acquiring key data from production equipment and sending the key data and the equipment's unique identifier to a first server; the first server analyzing the production status based on the key data and sending the analysis results to a second server; the second server determining whether the production equipment has experienced an anomaly based on the analysis results, and if so, sending the anomaly information to a user terminal; if no anomaly has occurred, the second server sending the analysis results to a third server; the third server predicting anomalies in the production equipment based on a target model; establishing anomaly patterns for each target model; performing in-depth analysis of the production equipment based on the anomaly patterns to obtain the type and cause of the anomaly, and sending this information to the user terminal; and the user terminal optimizing and adjusting the production process based on the analysis results. This application improves the production efficiency of aloe vera beverages.

Description

Technical Field

[0001] This application relates to the field of food production control technology, and in particular to production control methods and systems for aloe vera beverages. Background Technology

[0002] Aloe vera beverages, as a functional drink, are increasingly favored by consumers due to their health benefits and unique flavor. However, the production process of aloe vera beverages is relatively complex, involving several key stages. In this process, the identification of abnormal situations and the parameter control at each stage are crucial to the quality of the final product. Traditional aloe vera beverage production control methods mainly rely on manual experience and preset parameter settings.

[0003] To address the aforementioned issues, the food production industry has begun exploring intelligent production control methods. For example, Chinese patent application CN118466426A discloses an artificial intelligence-based optimization control method for the production process of fermented rice buns. Specifically, this method involves installing sensors to collect production characteristics and environmental optimization information during the fermented rice bun production process, transmitting this information to a database for storage and management. Historical data is used to train a machine learning model, analyzing the impact of data on fermentation effects, establishing a mathematical model to predict fermentation results, and providing feedback to production personnel. Based on monitoring data and prediction results, the raw material ratio is dynamically adjusted, introducing a closed-loop control system to continuously optimize the production status. A computer vision system is deployed to monitor production effects in real time and feed the results back to the production system. Combined with model analysis, this further optimizes production control, effectively preventing raw material waste caused by production anomalies and reducing production costs. It also allows for effective evaluation of production efficiency in cases of gas transfer or leakage from fermented rice containers. However, the aforementioned existing technologies primarily focus on optimizing the production process and do not explicitly explain how to handle abnormal situations.

[0004] For example, Chinese patent application CN110398907A discloses an automatic control method for meatball production. This method includes real-time acquisition of parameters associated with the meatballs and sending the acquired meatball images to a computer via an industrial camera assembly; receiving the meatball parameter comparison results calculated by the computer using a preset algorithm; and automatically controlling meatball production by activating the shaping and cooking structures through a meatball production CNC system when the comparison results match preset meatball state parameter values. This method uses multiple sensors to monitor the water environment of the meatballs, and the industrial camera assembly and computer calculate using a preset algorithm to automate meatball production. However, while the aforementioned prior art uses an industrial camera for real-time detection, it does not detail how to quickly respond to abnormal situations.

[0005] Therefore, a production control method and system for aloe vera beverages is needed to respond quickly to abnormal situations and optimize based on the cause of the abnormality. Summary of the Invention

[0006] This application provides a production control method and system for aloe vera beverages, which improves the production efficiency of aloe vera beverages.

[0007] In a first aspect, this application provides a production control method for aloe vera beverages, the method comprising:

[0008] Step S1: The sensor module acquires key data of the production equipment and sends the key data and the unique identifier of the production equipment itself to the first server;

[0009] Step S2: The first server analyzes the production status of the corresponding production equipment based on the unique identifier, obtains the first analysis result, and sends the first analysis result to the second server. The second server determines whether the production equipment is abnormal based on the first analysis result. If an abnormality occurs, the first analysis result is sent directly to the user terminal, and step S5 is executed.

[0010] Step S3: If no abnormality occurs, the second server sends the key data, the unique identifier, and the first analysis result to the third server. The third server includes multiple target models, and analyzes the key data and the first analysis result based on the target models to obtain the second analysis result.

[0011] Step S4: Establish an anomaly pattern for each target model. The third server performs in-depth analysis on the production equipment based on the anomaly pattern and the second analysis result to obtain a third analysis result, and sends the third analysis result to the user terminal.

[0012] Step S5: The user terminal optimizes and adjusts the production process based on the obtained first analysis result or the third analysis result.

[0013] In conjunction with the first aspect, in a first implementation of the first aspect of this application, the first server analyzes the production status of the corresponding production equipment based on the unique identifier, including:

[0014] Based on different data performance, the key data is divided into multiple data categories, including alarm data, parameter data, general data, and performance indicator data. Based on the data categories, the analysis of production status is divided into two implementation methods. For key data of the same category, the first server compares each data item in the key data with a preset threshold to determine whether it meets the normal range of the corresponding data category. If it does not meet the threshold, the status of the key data is determined to be abnormal.

[0015] For the key data of different categories, the first server combines the key data of different data categories based on the detection target to obtain multiple sample data, performs multi-dimensional analysis on the sample data, and obtains the analysis result of whether the production equipment is in an abnormal state.

[0016] In conjunction with the first aspect, in the second implementation of the first aspect of this application, directly sending the first analysis result to the user terminal further includes:

[0017] The abnormal states are divided into multiple abnormal types from simple to complex, including basic type, related type, predictable type and optimization type. An urgency level is set for each abnormal type, and the urgency level decreases progressively from simple to complex based on the abnormal type. Based on the urgency level, different notification methods are selected to send the first analysis results to the user terminal.

[0018] In conjunction with the first aspect, in a third implementation of the first aspect of this application, the analysis of the key data and the first analysis result based on the target model includes:

[0019] Key data from the normal production process of the aloe vera beverage are obtained and defined as historical normal data. Multiple variables are extracted from the historical normal data, and explanatory variables and target variables are defined from them. A prediction model for the target variable is established based on the explanatory variables. The prediction model is a mathematical model or a machine learning model. Each production stage includes one or more target models. The target model outputs the predicted value of the target variable. The current production stage of each key data point is identified based on a unique identifier. The corresponding target model is selected based on the production stage, and the key data is input into the corresponding target model to obtain the actual value of each target variable for each production stage. The actual value is defined as the second analysis result.

[0020] In conjunction with the first aspect, in the fourth implementation of the first aspect of this application, anomaly modes for each target model are established, including:

[0021] Based on each explanatory variable in the historical normal data, a perturbation method is determined. The perturbation method is to add outliers to each explanatory variable within a preset range while keeping other explanatory variables unchanged. The modified explanatory variables are combined with other explanatory variables to generate outlier data. Each outlier data is input into the corresponding target model. The target model outputs the predicted value of the target variable. The target variable value is compared with the predicted value of the historical normal data to obtain the deviation value for each target variable. All deviation values ​​are combined to obtain the anomaly pattern of each target model. Each target model has one or more anomaly patterns.

[0022] In conjunction with the first aspect, in the fifth implementation of the first aspect of this application, the third server performs in-depth analysis of the production equipment based on the abnormal mode and the second analysis result, including:

[0023] The deviation between the actual value of each target variable in the second analysis result and the predicted value obtained from historical normal data is calculated. Based on the combination of the deviation values, a deviation pattern of the current production equipment is generated, and the similarity between the deviation pattern and each abnormal pattern is calculated. The abnormal pattern with a similarity greater than a first threshold is selected as the current abnormal pattern. A fault prediction model is established based on a deep learning model, and fault data related to various abnormal patterns is obtained. The fault data includes fault type, fault occurrence time, and fault cause. The abnormal pattern and the corresponding fault data are input into the fault prediction model for training, and the current abnormal pattern is input into the trained fault prediction model to obtain the types of faults that may occur in the current abnormal pattern and the causes of the faults, which are defined as the third analysis result.

[0024] In conjunction with the first aspect, in the sixth implementation of the first aspect of this application, optimizing and adjusting the production process based on the obtained first analysis result or the third analysis result includes:

[0025] Record the optimization operations performed on the production process based on the first analysis result and the third analysis result, obtain the production state of the corresponding production stage after the optimization operation, and label the optimization operation. Build an optimization model based on reinforcement learning, input the optimization operation and the production state after the optimization operation into the optimization model, take the optimization operation as the action, take the production state as the reward, set a reward function based on the production goal, evaluate each action based on the reward function, obtain the optimal action, and execute the optimization operation based on the optimal action.

[0026] Secondly, this application provides a production control system for aloe vera beverages, the system comprising:

[0027] The acquisition module and sensor module acquire key data from the production equipment and send the key data and the unique identifier of the production equipment itself to the first server.

[0028] The anomaly detection module is as follows: the first server analyzes the production status of the corresponding production equipment based on the unique identifier, obtains a first analysis result, and sends the first analysis result to the second server. The second server determines whether the production equipment has an anomaly based on the first analysis result. If an anomaly occurs, the first analysis result is directly sent to the user terminal, and step S5 is executed.

[0029] If no anomaly occurs, the second server sends the key data, the unique identifier, and the first analysis result to the third server. The third server includes multiple target models, and analyzes the key data and the first analysis result based on the target models to obtain the second analysis result.

[0030] An anomaly analysis module establishes anomaly patterns for each target model. The third server performs in-depth analysis on the production equipment based on the anomaly patterns and the second analysis results to obtain a third analysis result, which is then sent to the user terminal.

[0031] The optimization module allows the user terminal to optimize and adjust the production process based on the obtained first analysis result or the third analysis result.

[0032] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0033] This invention avoids data confusion and ensures data traceability by assigning a unique identifier to each device. Sensor modules collect key data in real time and send it to a first server. The first server quickly completes basic status analysis; if an anomaly is detected, it directly notifies the user to make adjustments and reduce losses. When no anomaly is detected, the key data is sent to a second server. The second server defines a dedicated target model for each production stage and automatically matches the corresponding target model for each stage based on the unique identifier. The target model can predict changes in target variables based on key data in each stage. An anomaly pattern library corresponding to the target model is established. Combining current data with the pattern library, the probability and cause of current device failure are generated, providing early warning of potential failures and guiding planned maintenance. Based on the analysis results, the user terminal automatically obtains optimized actions using a reinforcement learning model, adapting to changes and uncertainties in the production process, reducing manual operation, and improving production efficiency. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of one embodiment of the production control method for aloe vera beverages in this application.

[0036] Figure 2 This is a flowchart illustrating the production status analysis in this application embodiment;

[0037] Figure 3 This is a schematic diagram illustrating the types of abnormal states and their corresponding urgency levels in the embodiments of this application.

[0038] Figure 4 This is a schematic diagram of one embodiment of the production control system for aloe vera beverages in this application. Detailed Implementation

[0039] This application provides a production control method and system for aloe vera beverages. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0040] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of a production control method for aloe vera beverages in this application includes:

[0041] Step S1: The sensor module acquires key data from the production equipment and sends the key data and the unique identifier of the production equipment to the first server.

[0042] Specifically, based on the production process of aloe vera beverages, the production equipment is divided into extraction equipment, mixing equipment, sterilization equipment, bottling equipment, and other auxiliary equipment. Each type of production equipment is assigned a unique identifier, which is associated with relevant equipment information, such as equipment name, model, and location. Key data from the production equipment, such as temperature and pressure data during the extraction stage, are collected in real time by sensors. The collected data is also preliminarily processed, and the key data and unique identifier are packaged into a data packet and sent to the first server via a selected communication method (such as Wi-Fi, Bluetooth, LPWAN, etc.). The first server is an edge server that can process the key data collected by the sensor module in real time and perform preliminary analysis locally, enabling the control system to respond quickly to abnormal situations in the production process.

[0043] Step S2: The first server analyzes the production status of the corresponding production equipment based on the unique identifier, obtains the first analysis result, and sends the first analysis result to the second server. The second server determines whether the production equipment has an abnormality based on the first analysis result. If an abnormality occurs, the first analysis result is sent directly to the user terminal, and step S5 is executed.

[0044] Specifically, such as Figure 2 The diagram shows a flowchart of production status analysis. The first server receives data packets from the sensor module, identifies the corresponding production equipment based on the unique identifier in the data packet, performs a preliminary analysis of the production status of the production equipment, and generates a first analysis result. The first analysis result includes, for example, the operating efficiency of the equipment and whether the operating parameters of the equipment are within the normal range. If they are outside the normal range, the production equipment is determined to be in an abnormal state. The second server receives the first analysis result from the first server. The second server is a cloud server and is mainly responsible for comprehensive anomaly judgment and user notification.

[0045] Based on the first analysis result, the cloud server determines whether the production equipment has malfunctioned. If an malfunction occurs, the first analysis result is sent directly to the user terminal, where the specific information of the malfunction, such as the equipment name and the parameter information of the malfunction, is displayed. Then, step S5 is executed, which will be explained in detail later.

[0046] Step S3: If no anomaly occurs, the second server sends the key data, unique identifier, and first analysis result to the third server. The third server includes multiple target models, and analyzes the key data and first analysis result based on the target models to obtain the second analysis result.

[0047] Specifically, the third server is an analysis server responsible for further data analysis and optimization. Multiple target models are pre-configured within the third server, with different target models corresponding to each production stage. Based on a unique identifier, the production equipment and its production stage are identified, and the corresponding target model is selected. For example, if the unique identifier corresponds to the extraction equipment, the target model for the extraction stage is selected, such as the extraction rate prediction model. Assuming key data such as extraction temperature = 85°C, extraction pressure = 5.5 bar, and extraction time = 2.5 hours, the first analysis result is: "The extraction equipment is operating normally." The key data and the first analysis result are input into each target model. Each target model performs in-depth analysis of the input data to predict the extraction rate of aloe vera from the raw materials in the extraction stage. The predicted values ​​of the extraction rate and other target variables are used as the second analysis result.

[0048] Step S4: Establish the anomaly pattern for each target model. The third server performs in-depth analysis on the production equipment based on the anomaly pattern and the second analysis result, obtains the third analysis result, and sends the third analysis result to the user terminal.

[0049] Specifically, the third server identifies potential anomalies in the current production state based on the second analysis results, matches potential anomalies in the current equipment from the anomaly patterns, predicts the probability and duration of equipment failure based on the anomaly patterns, analyzes the causes of failure, and obtains the third analysis results, such as "the predicted probability of equipment failure is 25%, the expected failure time is within 24 hours, and the cause of failure may be sensor failure or equipment wear."

[0050] Step S5: The user terminal optimizes and adjusts the production process based on the obtained first or third analysis results.

[0051] Specifically, based on the received analysis results, which may be the first analysis result (displaying information about equipment malfunctions) or the third analysis result (predicting which equipment may be malfunctioning and the cause of the malfunction), the user terminal adjusts the production equipment according to the analysis results. During the adjustment process, a reinforcement learning model is set up. Since the equipment status and environmental conditions in the production process may change over time, the reinforcement learning model can dynamically adjust the parameters of the production equipment based on real-time data, thereby ensuring that the production process is always in an optimal state. By evaluating the effect of the adjustment actions in real time, the reinforcement learning model can quickly find the optimal adjustment strategy, reducing waste and downtime in the production process.

[0052] This invention avoids data confusion and ensures data traceability by assigning a unique identifier to each device. Sensor modules collect key data in real time and send it to a first server. The first server quickly completes basic status analysis; if an anomaly is detected, it directly notifies the user to make adjustments to reduce losses. When no anomaly is detected, the key data is sent to a second server. The second server defines a dedicated target model for each production stage and automatically matches the corresponding target model for each stage based on the unique identifier. The target model can predict changes in target variables based on key data in each stage. An anomaly pattern library corresponding to the target model is established. Combining current data with the pattern library, the probability and cause of current device failure are generated, providing early warning of potential failures and guiding planned maintenance. Based on the analysis results, the user terminal automatically obtains optimized actions using a reinforcement learning model, adapting to changes and uncertainties in the production process, reducing manual operation, and improving production efficiency.

[0053] In one specific embodiment, the first server analyzes the production status of the corresponding production equipment based on a unique identifier, including the following steps:

[0054] Based on different data performance, key data is divided into multiple data categories, including alarm data, parameter data, general data, and performance indicator data. Based on the data category, the analysis of production status is divided into two implementation methods. For key data of the same category, the first server compares each data item in the key data with a preset threshold to determine whether it meets the normal range of the corresponding data category. If it does not meet the threshold, the key data is determined to be in an abnormal state.

[0055] For different categories of key data, the first server combines key data from different data categories based on the detection target to obtain multiple sample data. The sample data is then analyzed in multiple dimensions to obtain the analysis results on whether the production equipment is in an abnormal state.

[0056] Specifically, alarm data indicates equipment malfunctions or abnormalities, such as excessively high temperature sensor readings; parameter data indicates real-time data of key parameters during production, such as extraction temperature, sterilization temperature, extraction pressure, extract flow rate, and stirring speed; general data indicates data related to production plans, product quality, and equipment status, such as planned output and planned running time; and performance index data is used to evaluate production efficiency and equipment performance, such as hourly output and cumulative output.

[0057] The received key data is divided into different data categories. For the same type of data, assuming the data category is parameter data and the data item includes extraction temperature, the actual extraction temperature is compared with the preset normal range (e.g., 40°C to 50°C) to determine whether it is within the normal range. If the extraction temperature is 60°C, which does not meet the normal range, the first analysis result containing the temperature abnormality in the extraction stage is directly sent to the user terminal, and the user terminal can respond to the abnormality quickly in a short time.

[0058] However, a single data category may not fully reflect the complexities of the production process. This invention also provides a second method for judging abnormal states in the production process. Key data from different categories are combined according to the detection target. For example, if the detection target is whether production equipment frequently stops briefly, general data 1: the running time of production equipment A in the past 10 minutes; performance index data 1: the number of brief stops of production equipment A, for example, 2 times; combined to obtain sample data 1: the number of brief stops of production equipment A in the past 10 minutes is 2. This number is compared with the number in the past 10 minutes. If the number of brief stops increases by more than 50%, the production equipment is judged to be abnormal. Another example: if the target is to analyze whether the number of defective products has increased abnormally, general data 2: the number of defective products is 5; performance index data 1: inspection time (for example, in the past hour); combined to obtain sample data 2: the number of defective products in the past hour is 5. Sample data 2 is compared with the number at the previous inspection. If the increase in defective product data exceeds a preset threshold, it is judged to be an abnormal state.

[0059] There may be correlations between different categories of data. For example, equipment uptime (general data) may affect production efficiency (performance index data), while changes in temperature and pressure (parameter data) may affect product quality. This invention, by comprehensively analyzing different categories of data, can provide a more holistic assessment of the equipment's operating status.

[0060] In one specific embodiment, directly sending the first analysis result to the user terminal further includes the following steps:

[0061] Abnormal states are categorized into multiple anomaly types, ranging from simple to complex, including basic types, related types, types to be predicted, and types to be optimized. An urgency level is set for each anomaly type, with the urgency level decreasing progressively from simple to complex. Based on the urgency level, different notification methods are selected to send the first analysis results to the user terminal.

[0062] Specifically, such as Figure 3The diagram illustrates the types of abnormal states and their corresponding urgency levels. Abnormal states are categorized into four types: Basic types: involving basic equipment operating status, such as equipment downtime and simple alarms; Correlated types: involving correlation analysis of multiple data categories, such as the relationship between temperature and production efficiency; Predictable types: involving anomalies predicted by algorithms, such as the prediction of production end time (this will be explained in more detail later); and Optimizable types: involving the causes of anomalies and optimization suggestions, such as production plan optimization. The urgency of these four types decreases from highest to lowest: Basic types: highest urgency (e.g., equipment downtime, requiring immediate action); Correlated types: medium urgency (e.g., decreased production efficiency, requiring timely adjustments); Predictable types: low urgency (e.g., production end time prediction, requiring advance preparation); Optimizable types: lowest urgency (e.g., production plan optimization, which can be handled later).

[0063] Depending on the urgency of the anomaly type, different notification methods are selected to send the initial analysis results to the user terminal. For example, basic anomalies are notified via monitors on the production floor, prompting operators to immediately check the equipment; related anomalies are sent to the user terminal, prompting operators to adjust equipment parameters; anomalies to be predicted are sent to the user terminal and relevant management personnel are notified via email; and anomalies to be optimized are notified via email to the production planner for subsequent optimization.

[0064] In one specific embodiment, the key data and the first analysis results are analyzed based on the target model:

[0065] Key data from the normal production process of aloe vera beverages are obtained and defined as historical normal data. Multiple variables are extracted from the historical normal data, and explanatory and target variables are defined from them. A predictive model for the target variable is built based on the explanatory variables. The predictive model can be a mathematical model or a machine learning model. Each production stage includes one or more target models. The target model outputs the predicted value of the target variable. The current production stage of each key data is identified based on a unique identifier. The corresponding target model is selected based on the production stage, and the key data is input into the corresponding target model to obtain the actual value of each target variable for each production stage. The actual value is defined as the second analysis result.

[0066] Specifically, historical normal production data of aloe vera beverages is obtained from a historical database. This data includes key data for each production stage. Multiple variables are extracted from this historical normal data, such as: Extraction stage: extraction temperature (x1), extraction pressure (x2), extraction time (x3), extraction rate (y1); Mixing stage: stirring speed (x4), stirring time (x5), raw material ratio (x6), mixing uniformity (y2); Sterilization stage: sterilization temperature (x7), sterilization time (x8), product pH value (x9), microbial inactivation rate (y3). The variables to be predicted are defined as follows: The target variable is such as extraction rate, mixing uniformity, and microbial inactivation rate. Variables affecting the target variable are used as explanatory variables, such as extraction temperature and extraction pressure. Based on the explanatory variables, a predictive model for the target variable is constructed. The predictive model can be a mathematical model (e.g., multiple linear regression) or a machine learning model (e.g., neural network, support vector machine). Each production stage can contain one or more target models. For example: extraction stage: one target model is used to predict the extraction rate; mixing stage: two target models are used to predict mixing uniformity and stirring power, respectively; sterilization stage: one target model is used to predict the microbial inactivation rate.

[0067] Assuming the target model is a mathematical model, the extraction rate prediction model is: extraction rate = f(extraction temperature, extraction pressure, extraction time). The second server identifies the production stage based on the unique identifier, selects the corresponding target model, and inputs the current key data of the production equipment into the target model for predictive analysis. It outputs the predicted value of the target variable, which is defined as the second analysis result, for example, the predicted extraction rate is 95%.

[0068] In one specific embodiment, establishing the anomaly pattern for each target model includes the following steps:

[0069] The perturbation method is determined for each explanatory variable in the historical normal data. The perturbation method is to add outliers to each explanatory variable within a preset range while keeping other explanatory variables unchanged. The changed explanatory variables are combined with other explanatory variables to generate outlier data. Each outlier data is input into the corresponding target model. The target model outputs the predicted value of the target variable. The target variable value is compared with the predicted value of the historical normal data to obtain the deviation value for each target variable. All deviation values ​​are combined to obtain the outlier pattern of each target model. Each target model has one or more outlier patterns.

[0070] Specifically, for each explanatory variable, the perturbation method is determined. Assuming the explanatory variable is extraction temperature, the corresponding perturbation method is to increase the extraction temperature by 5°C while keeping other explanatory variables unchanged. The perturbation range can be determined based on the actual situation. If the explanatory variable has a small influence on the target variable, a larger perturbation range can be set to observe abnormal changes in the target variable. If the explanatory variable has a large influence on the target variable, the target variable can be changed within a smaller perturbation range. Each perturbed explanatory variable is combined with other unperturbed explanatory variables to generate outlier data. For example, outlier data 1: [extraction temperature +5°C, extraction pressure, extraction time, stirring speed, stirring time].

[0071] Each abnormal data point is input into the corresponding target model, namely the extraction rate prediction model. The target model outputs the predicted target variable value. The predicted value is compared with the predicted value of historical normal data to calculate the deviation value. All deviation values ​​are combined to form the abnormality pattern for each target model. For example, abnormality pattern 1: [-5%, -3%, +2%], corresponding to a 3% decrease in extraction rate, a 3% decrease in mixing uniformity, and a 2% increase in microbial inactivation rate, respectively. Each target model can generate one or more abnormal patterns to comprehensively cover various abnormal situations. By identifying abnormal patterns, potential problems can be predicted in advance, the causes of abnormalities can be accurately identified, production parameters can be optimized, human intervention can be reduced, and the stability of the production process can be improved.

[0072] In one specific embodiment, the third server performs in-depth analysis of the production equipment based on the anomaly pattern and the second analysis results, specifically including the following steps:

[0073] The deviation between the actual value of each target variable in the second analysis result and the predicted value obtained from historical normal data is calculated. Based on the combination of deviation values, the deviation pattern of the current production equipment is generated, and the similarity between the deviation pattern and each abnormal pattern is calculated. The abnormal pattern with a similarity greater than the first threshold is selected as the current abnormal pattern. A fault prediction model is established based on a deep learning model, and fault data related to multiple abnormal patterns is obtained. The fault data includes fault type, fault occurrence time, and fault cause. The abnormal pattern and the corresponding fault data are input into the fault prediction model for training. The current abnormal pattern is input into the trained fault prediction model to obtain the types of faults that may occur in the current abnormal pattern and the causes of the faults, which are defined as the third analysis result.

[0074] Specifically, a deviation pattern for the current production equipment is generated based on the deviation value and matched with abnormal patterns in a predefined abnormal pattern library. The average difference of deviation values ​​for each category is calculated; the smaller the average, the greater the similarity. Abnormal patterns with similarity greater than a first threshold are defined as abnormal patterns for the current production equipment. The second server selects the corresponding fault prediction model based on the identified abnormal pattern, for example, abnormal pattern 1: [-5%, -3%, +2%]. It obtains fault data related to this abnormal pattern, including fault type (e.g., equipment failure), fault occurrence time (occurred 3 times in the past 6 months), and fault cause (temperature sensor failure, pressure sensor failure). The fault data is input into the fault prediction model for training. The current abnormal pattern is input into the trained fault prediction model to obtain possible fault types and causes. The user terminal sends the fault information to the user terminal, displaying a fault warning (e.g., "Equipment fault warning: Temperature sensor failure may occur") and a fault cause analysis report (e.g., "The fault cause may be aging or damage to the temperature sensor").

[0075] In one specific embodiment, optimizing and adjusting the production process based on the obtained first or third analysis results specifically includes the following steps:

[0076] Record the optimization operations performed on the production process based on the first and third analysis results, obtain the production status of the corresponding production stage after the optimization operation, and label the optimization operation. Build an optimization model based on reinforcement learning, input the optimization operation and the production status after the optimization operation into the optimization model, take the optimization operation as the action and the production status as the reward, set a reward function based on the production goal, evaluate each action based on the reward function, obtain the optimal action, and execute the optimization operation based on the optimal action.

[0077] Specifically, the user terminal receives the first or third analysis result from the first or third server. The user makes adjustments based on the equipment in the analysis result that is in an abnormal state. However, the range of parameters may not be particularly determined when making adjustments. This invention establishes a reinforcement learning model based on a deep Q-network (DQN). The reinforcement learning model defines the current production state as a state, with a state vector of [extraction temperature, extraction pressure, stirring speed, product quality index, equipment state]. The adjusted parameters or the replacement of equipment are defined as actions. The user terminal designs a reward function based on the production goal to evaluate the quality of each action.

[0078] For example, if the objective is to increase the extraction rate while reducing energy consumption, the corresponding reward = w1 × (actual extraction rate - target extraction rate) / target extraction rate × 100% - w2 × (actual energy consumption - target energy consumption) / target energy consumption × 100%, where w1 and w2 are weighting coefficients. The reinforcement learning model evaluates the expected cumulative reward for each action based on the current state and the reward function, and selects the optimal action. For example, increasing the extraction temperature by 1°C has the highest expected cumulative reward, so increasing the extraction temperature by 1°C is selected as the optimal action. The reinforcement learning-based optimization and adjustment method can significantly improve the automation adjustment efficiency of the aloe vera beverage production process, adapt to changes and uncertainties in the production process, reduce manual operation, and improve production efficiency.

[0079] The above describes a production control method for aloe vera beverages according to embodiments of this application. The following describes a production control system for aloe vera beverages according to embodiments of this application. Please refer to [link / reference]. Figure 4 One embodiment of a production control system for aloe vera beverages in this application includes:

[0080] The acquisition module and sensor module acquire key data from the production equipment and send the key data and the unique identifier of the production equipment to the first server;

[0081] The anomaly detection module involves the first server analyzing the production status of the corresponding production equipment based on a unique identifier, obtaining a first analysis result, and sending the first analysis result to the second server. The second server determines whether an anomaly has occurred in the production equipment based on the first analysis result. If an anomaly has occurred, the first analysis result is directly sent to the user terminal, and step S5 is executed.

[0082] If no anomaly occurs, the second server sends the key data, unique identifier, and first analysis result to the third server. The third server includes multiple target models, which analyze the key data and first analysis result based on the target models to obtain the second analysis result.

[0083] The anomaly analysis module establishes anomaly patterns for each target model. The third server performs in-depth analysis of the production equipment based on the anomaly patterns and the second analysis results, obtains the third analysis results, and sends the third analysis results to the user terminal.

[0084] The optimization module allows the user terminal to optimize and adjust the production process based on the obtained first or third analysis results.

[0085] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0086] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A production control method for aloe vera beverages, characterized in that, The method includes: Step S1: The sensor module acquires key data of the production equipment and sends the key data and the unique identifier of the production equipment itself to the first server; Step S2: The first server analyzes the production status of the corresponding production equipment based on the unique identifier, obtains a first analysis result, which is the analysis result of whether the production equipment is in an abnormal state, and sends the first analysis result to the second server. The second server determines whether the production equipment is abnormal based on the first analysis result. If an abnormality occurs, the first analysis result is directly sent to the user terminal, and step S5 is executed. Step S3: If no abnormality occurs, the second server sends the key data, the unique identifier, and the first analysis result to the third server. The third server includes multiple target models. Based on the target models, the key data and the first analysis result are analyzed to obtain a second analysis result. The second analysis result is the predicted value of each target variable for each production stage. Step S4: Establish an anomaly pattern for each target model. The third server performs in-depth analysis on the production equipment based on the anomaly pattern and the second analysis result to obtain a third analysis result, and sends the third analysis result to the user terminal. The third analysis result is the type of failure that may occur in the current anomaly pattern and the cause of the failure. Step S5: The user terminal optimizes and adjusts the production process based on the obtained first analysis result or the third analysis result.

2. The method according to claim 1, characterized in that, The first server analyzes the production status of the corresponding production equipment based on the unique identifier, including the following steps: Based on different data performance, the key data is divided into multiple data categories, including alarm data, parameter data, general data, and performance indicator data. Based on the data categories, the analysis of production status is divided into two implementation methods. For key data of the same category, the first server compares each data item in the key data with a preset threshold to determine whether it meets the normal range of the corresponding data category. If it does not meet the threshold, the status of the key data is determined to be abnormal. For the key data of different categories, the first server combines the key data of different data categories based on the detection target to obtain multiple sample data, performs multi-dimensional analysis on the sample data, and obtains the analysis result of whether the production equipment is in an abnormal state.

3. The method according to claim 2, characterized in that, Sending the first analysis result directly to the user terminal also includes the following steps: The abnormal states are divided into multiple abnormal types from simple to complex, including basic type, related type, predictable type and optimization type. An urgency level is set for each abnormal type, and the urgency level decreases progressively from simple to complex based on the abnormal type. Based on the urgency level, different notification methods are selected to send the first analysis results to the user terminal.

4. The method according to claim 1, characterized in that, The analysis of the key data and the first analysis result based on the target model includes the following steps: Key data from the normal production process of the aloe vera beverage are obtained and defined as historical normal data. Multiple variables are extracted from the historical normal data, and explanatory variables and target variables are defined from them. A prediction model for the target variable is established based on the explanatory variables. The prediction model is a mathematical model or a machine learning model. Each production stage includes one or more target models. The target model outputs the predicted value of the target variable. The current production stage of each key data point is identified based on a unique identifier. The corresponding target model is selected based on the production stage, and the key data is input into the corresponding target model to obtain the predicted value of each target variable for each production stage. The predicted value is defined as the second analysis result.

5. The method according to claim 4, characterized in that, Establishing anomaly patterns for each target model involves the following steps: Based on each explanatory variable in the historical normal data, a perturbation method is determined. The perturbation method is to add outliers to each explanatory variable within a preset range while keeping other explanatory variables unchanged. The modified explanatory variables are combined with other explanatory variables to generate outlier data. Each outlier data is input into the corresponding target model. The target model outputs the predicted value of the target variable. The target variable value is compared with the predicted value of the historical normal data to obtain the deviation value for each target variable. All deviation values ​​are combined to obtain the anomaly pattern of each target model. Each target model has one or more anomaly patterns.

6. The method according to claim 5, characterized in that, The third server performs in-depth analysis of the production equipment based on the anomaly pattern and the second analysis result, including the following steps: The deviation between the actual value of each target variable in the second analysis result and the predicted value obtained from historical normal data is calculated. Based on the combination of the deviation values, a deviation pattern of the current production equipment is generated, and the similarity between the deviation pattern and each abnormal pattern is calculated. The abnormal pattern with a similarity greater than a first threshold is selected as the current abnormal pattern. A fault prediction model is established based on a deep learning model, and fault data related to various abnormal patterns is obtained. The fault data includes fault type, fault occurrence time, and fault cause. The abnormal pattern and the corresponding fault data are input into the fault prediction model for training, and the current abnormal pattern is input into the trained fault prediction model to obtain the types of faults that may occur in the current abnormal pattern and the causes of the faults, which are defined as the third analysis result.

7. The method according to claim 1, characterized in that, Optimizing and adjusting the production process based on the obtained first or third analysis results includes the following steps: Record the optimization operations performed on the production process based on the first analysis result and the third analysis result, obtain the production state of the corresponding production stage after the optimization operation, and label the optimization operation. Build an optimization model based on reinforcement learning, input the optimization operation and the production state after the optimization operation into the optimization model, take the optimization operation as the action, take the production state as the reward, set a reward function based on the production goal, evaluate each action based on the reward function, obtain the optimal action, and execute the optimization operation based on the optimal action.

8. A production control system for aloe vera beverages, used to implement the production control method for aloe vera beverages as described in any one of claims 1-7, characterized in that, The system includes: The acquisition module and sensor module acquire key data from the production equipment and send the key data and the unique identifier of the production equipment itself to the first server. The anomaly detection module is as follows: the first server analyzes the production status of the corresponding production equipment based on the unique identifier, obtains a first analysis result, which is the analysis result of whether the production equipment is in an abnormal state, and sends the first analysis result to the second server. The second server determines whether the production equipment has an anomaly based on the first analysis result. If an anomaly has occurred, the first analysis result is directly sent to the user terminal, and step S5 is executed. If no anomaly occurs, the second server sends the key data, the unique identifier, and the first analysis result to the third server. The third server includes multiple target models. Based on the target models, the key data and the first analysis result are analyzed to obtain a second analysis result. The second analysis result is the predicted value of each target variable for each production stage. An anomaly analysis module establishes anomaly patterns for each target model. The third server performs in-depth analysis on the production equipment based on the anomaly patterns and the second analysis results to obtain a third analysis result, which is then sent to the user terminal. The third analysis result represents the types of faults that may occur in the current anomaly pattern and the causes of the faults. The optimization module allows the user terminal to optimize and adjust the production process based on the obtained first analysis result or the third analysis result.

Images